Productivity Effects of Business Process Outsourcing (BPO)A Firm-level Investigation Based on Panel Data∗

Jörg Ohnemus†

Centre for European Economic Research (ZEW), Mannheim

April 2009

Work In Progress, Please Do Not Cite!

Abstract

This paper analyses the impact of business process outsourcing (BPO) on firm pro-ductivity based on a comprehensive German firm-level panel data set covering man-ufacturing and service industries. The growing importance of service inputs intothe production process is undisputed. Firms increasingly decide to go to the mar-ket and buy all or at least parts of selected services they need from external serviceproviders. This is especially true for services which rely to great extend on newinformation and communication technologies. Doing so, outsourcing firms can con-centrate on their core competencies. Additionally, they benefit from the expertise ofthe external service provider. Finally, external vendors are able to provide servicesat lower price because of scale effects. By estimating a production function using asystem-generalised method of moments (system-GMM) estimation procedure, I showthat BPO has a positive and highly significant impact on firm-level productivity inboth manufacturing and service industries.

Keywords: Business Process Outsourcing (BPO), Productivity, Panel Data, System-GMMJEL-Classification: C23, D24, L24, L60, L80

∗ This paper was written as part of the research project “Business Process Outsourcing: Determinants andConsequences for Competitiveness” commissioned by the Förderkreis Wissenschaft und Praxis at the Centrefor European Economic Research.

† Centre for European Economic Research (ZEW), Research Group Information and Communication Technolo-gies, P.O. Box 10 34 43, D–68034 Mannheim, Germany, Tel.: +49–(0)621–1235–354, Fax: +49–(0)621–1235–4354, ohnemus@zew.de.

1 Introduction

The overall importance of services as an input into the production process of firms is undisputed.

Firms can choose between two different forms of acquiring those inputs: they can produce

services themselves or they can contract out the services to an external service provider.1 Figure

1 reflects the growing importance of external service inputs for the production process of the

German economy. Between 1995 and 2005 (the latest point of time for which input-output data

is available) the share of intermediate inputs from the “other business activities sector” (NACE

74) in the total production value in Germany increased from 5.45 percent to 6.38 percent.

Although this increase in percentage terms seems rather moderate, the absolute numbers are

quite substantial. For the period from 1997 to 2005 the increase in the share of intermediate

input of 1.3 percentage points reflects an additional amount of business services requested

by the market of around e91.5 billion. The import share only represents an additional e3.9

billion. The rest of the increase therefore comes from service providers located in Germany.

Since concentrating solely on the “other business activities sector” is quite narrow when focusing

on business process outsourcing activities, a wider definition which includes additional sectors

providing business services is displayed in Figure 2. Here I show results for the “corporate service

sector” which comprises the sectors “computer and related activities” (NACE 72), “research

and development” (NACE 73) and “other business activities” (NACE 74). In this case, the

increasing importance of services as inputs to the production process is even more pronounced.

Between 1997 and 2005 (again the latest point in time with input-output data availability) the

increase of services from the above defined sectors as inputs into other sectors increased by 1.5

percentage points to a total amount of almost e307 billions (including imports of around e24

billions). Data made available by the EU-Klems project allows to take a look at the share of

intermediate service inputs in total value added in Germany of the past years until the year

1978. Since then the share of service input rose from 24.8 percent to 34.0 percent in 2005 for

the service sector, an increase of more than 9 percentage points. Regarding the manufacturing

industries, the increase is smaller but still amounts to almost 6 percentage points in the same

time span. A study conducted by the Centre of European Economic Research (ZEW) in 2007

1 Sometimes firms, especially larger ones and those with several subsidiaries, found their own service divisionwhich then provides services to all the other parts of the group. Sometimes those service divisions also beginto offer their service to external companies.

1

analyses the involvement of German firms in business process outsourcing. They find that

54 percent of German manufacturing firms and 59 percent of firms belonging to the service

industries are involved in BPO (ZEW, 2007). This holds especially in the fields of accounting

and human resource managements where the services of BPO providers are highly demanded by

firms in both industries (see Figure 4).

The aforementioned figures illustrate the growing importance of external service provision for

the German economy. The question now is to find out if this usage also has an effect on the

business performance of the outsourcing firm, e.g. in terms of productivity gains. Why should

firms resort to external providers and give away decision power and (maybe) flexibility if it does

not help them to improve their economic performance in some way? Therefore, the purpose of

this paper is to analyse the impact of business process outsourcing on firm-level productivity by

using a comprehensive panel data set from the German manufacturing and service industries

for the years 2000, 2002, 2004 and 2007. I apply different estimation methods to a production

function framework, among others system-GMM estimation. This controls for unobserved firm

effects, measurement errors in the variables and simultaneity of inputs and output, which may

induce substantial biases in pooled OLS regressions. The results clearly show a positive and

highly significant impact of business process outsourcing on the productivity of outsourcing firms.

The remainder of the paper is structured as follows: Section 2 gives a definition of business

process outsourcing and develops the main hypothesis. Furthermore, an empirical literature

review focusing on business process outsourcing and productivity research is presented. Section

3 introduces the estimation procedure. In Section 4 the data set and the applied transformation

steps are presented. Section 5 discusses the estimation results. Section 6 concludes.

2

2 Background Information

2.1 Business Process Outsourcing

Business Process Outsourcing (BPO) is a broad term referring to outsourcing in all fields of

economic activity. It can be defined as ‘an organisation entering into a contract with another

organisation to operate and manage one or more of its business processes or sub-processes’

(Sharma, 2004). Focusing on the more technical side of BPO an alternative definition would

be ‘BPO is the leveraging of technology or specialist process vendors to provide and manage

an organisation’s critical and/or non-critical enterprise processes and applications’ (Mukherjee,

2007). A BPO service provider differentiates itself from a typical third party Application

Service Provider (ASP) by either putting in new technology or applying existing technology in

a new way to improve a (business) process. The application of BPO usually also means that

a certain amount of risk is transferred to the vendor which runs the process on behalf of the

outsourcer. Business segments which are typically outsourced include information technology

(IT), human resources, facilities and real estate management, and accounting. In recent years,

because of the rapid development of information technology, business process outsourcing has

increasingly become the delegation of one or more IT-intensive business processes to an external

provider that in turn owns, administers and manages the selected process based on defined

and measurable performance criteria. With BPO, the outsourcing provider not only takes over

administrative responsibility for a technical function but also assumes strategic responsibility for

the execution of a complete, business-critical function. This additional step can introduce new

efficiencies and cost savings while it also enables the outsourcing vendor to deliver important

strategic benefits to the customer. BPO is positively related to the search for more efficient

organisational designs: cost reduction, productivity growth and innovative capabilities. Hence,

a source of strategic advantage.

An important fact of business process outsourcing is its ability to free corporate executives from

some of their day-to-day process management responsibilities. Executives usually spend most

of their time managing everyday business and only some time on formulating strategies for a

successful advancement of the company. This can look quite different when business processes

are outsourced. Once a process is successfully outsourced, the ratio can be easily reversed. As

3

a result, executives have more time. This saved time helps them tremendously to explore new

revenue areas, accelerate other projects and focus on their customers. This, beyond any doubt,

leads to productivity improvements.

Companies that outsource their business processes are often able to generate new efficiencies and

improve their productivity. They are in a better position to reallocate their resources to other

important projects. This also helps their employees to increase their efficiency and productivity.

In most cases highly qualified experts are brought in to design and manage these processes.

Those experts bring with them increased productivity and years of experience that most

companies previously did not have access to or could not afford on their own. The availability

of highly qualified skill pools and the faster adoption of well-defined business processes leads to

productivity improvements without compromising on quality.

2.2 Outsourcing of Services

Business Process Outsourcing is very closely related to service outsourcing since it can be seen as

the outsourcing activity of a sub-group of services. The empirical literature dealing with service

outsourcing and offshoring (the outsourcing to a service provider abroad) is much diversified.

Some papers deal with the outsourcing drivers, e.g. with the factors that affect the outsourcing

decision of the firm (Abraham and Taylor, 1996; Girma and Görg, 2004; Antonietti and Cainelli,

2007; Merino and Rodríguez Rodríguez, 2007).2 Another topic the empirical literature on

service outsourcing is dealing with, is the effect of outsourcing on firm success, e.g. in terms of

productivity increases of the outsourcing and/or offshoring firms. Important contributions were

made by Siegel and Griliches (1992); Fixler and Siegel (1999); Görg and Hanley (2005); Jiang

and Qureshi (2006); Sako (2006); Hijzen (2007) and Görg et al. (2008). All those papers analyse

the relationship between productivity and service outsourcing in general. Heshmati (2003) and

Olsen (2006) provide surveys of the more general literature on outsourcing and its relationship

to efficiency and productivity. The extent to which services are outsourced is treated by the

paper of Amiti and Wei (2005). They find that the international outsourcing of material inputs

2 Theoretical aspects about the determinants of outsourcing and offshoring can be found in Grossman andHelpman (2003, 2005) and Antràs et al. (2006)

4

is still more important than the international outsourcing of services. However, taking a look on

the dynamics of both outsourcing aspects, the growth rate of service outsourcing is much higher

compared to the growth rate of material outsourcing.

2.3 Outsourcing and ICT

Business process outsourcing is very closely related to information technology outsourcing. BPO

even comprises IT outsourcing to some extend. The increasing importance of information and

communication technologies (ICT), especially the usage of computers and intra and internet

network connections, has revolutionised the provision of services. Even more important, a

broad variety of new services were created because of the possibilities offered by new and

fast developing information and communication technologies. Therefore, the vast majority of

business processes today rely in some way on information and communication technologies. In

addition, the operation of a firm’s ICT infrastructure itself can be interpreted as business process.

Various authors have analysed ICT/IT outsourcing and offshoring, including Loh and Venkatra-

man (1992) as well as Barthêlemy and Geyer (2001; 2004; 2005), who took a close look on the

determinants of IT outsourcing. Additionally, further research was devoted to the outsourcing

firms’ performance, basically trying to identify (labour) productivity effects of IT outsourcing.

Maliranta et al. (2008) thereby find out that IT outsourcing enhances an organisation’s IT use

and thus boosts its labour productivity. In contrast, Bertschek and Müller (2006) cannot find

any significant differences in key variables between outsourcing and non-IT outsourcing firms.

They even find that firms without IT outsourcing produce more efficiently than those involved

in IT outsourcing. Ohnemus (2007) in turn finds the opposite. He shows that IT outsourcing

firms are more efficient in their production processes. Furthermore, he finds that employees

working at a computerised workplace are more productive in IT outsourcing firms. Besides the

empirical literature dealing with IT outsourcing, there is also a variety of theoretical papers

5

analysing various aspects of IT outsourcing. For a comprehensive overview see Dibbern et al.

(2004).3

The analysis of Abramovsky and Griffith (2006) rather attempts to identify the impact of

information and communication technologies on the make or buy decision of business services.

They rely on British firm-level data for the years 2001 and 2002. Information and commu-

nication technologies reduce the transaction costs and the costs of adjustment of outsourcing

relationships. Therefore, long distance service outsourcing (offshoring) has become increasingly

popular during the last few years. Additionally, the costs of these outsourcing activities

decreased significantly. Abramovsky and Griffith (2006) find out that ICT-intensive firms

tend to buy a greater amount of services on the market (compared to in-house production).

Furthermore, these firms are more involved in international service outsourcing (offshoring).

3 Empirical Framework

In order to estimate the impact of business process outsourcing on a firm’s productivity, I refer

to a production function framework using a Cobb-Douglas specification:

Yit = F (Ait, Lit,Kit, BPOi)

= Ait Lαit Kβit eγBPOi , (1)

where Yit refers to output of firm i at time t, Lit represents labour input and Kit represents

capital input. BPOi is a dummy variable indicating if firm i has been involved in business

process outsourcing since the year 2000 or before. After taking logs on both sides, one can

rewrite Equation (1) in the following way:

yit = α lit + β kit + γ BPOi + ηi + λj(i),t + εit, (2)

3 In their literature overview Dibbern et al. (2004) analyse 84 papers published between 1992 and 2000. Theyfind that most of the studies focus on Transaction Cost Theory, Agency Theory or Strategic ManagementTheory as a reference framework to explain IT outsourcing.

6

where small letters denote the corresponding logarithmic value and multifactor productivity

ln(Ait) = ηi + λj(i),t + εit is decomposed into a firm-specific and time-invariant fixed part

ηi, a time-variant industry-specific part λj(i),t,4 and a time-variant firm- specific residual εit.

Firm-effect ηi captures fixed or quasi-fixed factors affecting productivity, e.g. management

ability, organisational capital, branding or location. The residual εit comprises measurement

errors, mit, and firm-specific productivity shocks, µit, such that εit = mit + µit. In this anal-

ysis, bothmit and µit are assumed to be serially uncorrelated and only their sum εit is considered.

The industry time–variant part λj(i),t captures variations in productivity that are specific

to a particular industry and that are left unexplained by the input variables. In this sense,

λj(i),t helps to ensure that outputs of firms are more readily comparable across industries.

In particular, demand fluctuations induced by industry–specific business cycles may lead to

variations in factor utilisation that are similar across firms of one industry. The resulting in-

dustry–specific changes of productivity are then captured by λj(i),t. While the industry–specific

component λj(i),t will be controlled for by including time–variant industry dummies, distorting

effects from unobserved ηi and εit will be addressed by econometric techniques. I account for

the fact that both ηi and εit may be correlated with the inputs if, for example, firms with

a good management (i.e. a high ηi) are both more productive and more inclined to make

use of capital input, or if a demand shock (high εit) raises both productivity as well as investment.

The endogeneity of the explanatory variables can be removed by an instrumental variable

regression. In this respect, it is convenient to use GMM estimations with internal instruments,

i.e. other moments of the same variable (Hempell, 2006). More precisely, the first differences

of the explanatory variables are instrumented here by the levels of the lagged variables. The

prediction power of the internal instruments could be small, however, given the only minor

changes in some of the variables (e.g. number of employees) from one year to another. That

could evoke biases in the GMM estimator in first differences (Blundell and Bond, 1998).

Therefore I prefer the so-called system-GMM estimator of Arellano and Bover (1995). Here,

the differences are instrumented again with lagged levels as internal instruments. The levels

4 With j(i) denoting the industry j that firm i is operating in.

7

of the covariates are simultaneously instrumented by adequate lagged differences. The main

advantage of this approach is that besides the temporary differences, differences among firms

in levels are also taken account of in the estimation. That improves the information used for

identifying the effect and usually enhances the precision of the estimator. A necessary condition

for the system-GMM estimator is that the correlations between the unobserved fixed effects

and the covariates remain constant over time (Arellano and Bover, 1995). The productivity

estimations are carried using the command xtabond2 in STATA 10.1 (Roodman, 2006). I

applied the available two-step estimation variant which is asymptotically more efficient than the

one-step alternative. Unfortunately, the reported two-step standard errors tend to be severely

downwardly biased (Arellano and Bond, 1991; Blundell and Bond, 1998). To resolve this

problem, Windmeijer’s adjustment process for variances is applied (Windmeijer, 2005). This

method helps to make the two-step system-GMM estimation more efficient than the one-step

estimation.

4 Data

The firm-level data used for the empirical analysis are taken from a survey conducted by the

Centre for European Economic Research (ZEW) between 2000 and 2007. It is a representative

survey of information and communication technologies in the German manufacturing and

selected service sectors.5 In each wave a total of approximately 4,400 firms were surveyed.

The data is stratified according to industries (seven manufacturing industries and seven service

sectors), size (three distinct classes) and region (East or West Germany). Merging all four

existing waves of the collected data results in an (unbalanced) panel structure in important

key variables. Besides a huge amount of variables dealing with information and communication

technologies, the ZEW ICT-survey contains annual data on sales, number of employees (and their

skill structure) and expenditures on gross investment. In the last wave, which was conducted

in 2007, information about ICT and business process outsourcing was collected. The survey

5 The first wave of the so called ZEW ICT-survey was conducted in the year 2000, the second wave followed twoyears later, the third wave in 2004 and the hitherto last survey wave took place in 2007.

8

did not only focus on the present state of the outsourcing activities of each firm but also asked

for the starting year of various BPO activities (for a further discussion of this point, see page 11).

In order to conduct meaningful production function estimations, same of the available variables

have to be transformed using external data sources. In the following I will illustrate how this

external data is used to transform the available survey data which enables me to run production

function regressions. For output Yit I use a measure of firms’ value added. Alternatively, the

originally available data on total sales from the survey could be used with including firm–level

intermediate goods as an additional input. However, the available data lacks reliable information

on intermediate inputs. Using sales for output instead of value added without the inclusion of

the amount of intermediate inputs might lead to an omitted variable bias in the regressions since

industries that operate rather at the end of the value chain (such as wholesale and trade) resort

more to intermediate goods in terms of quantity than other industries do. In order to control

for these differences and to deflate the corresponding outputs, I calculate the shares of real value

added in nominal gross output at the NACE two–digit industry level.6 The firm–level data on

sales are then multiplied by these industry–specific shares.7 For labour input, the number of

employees in full-time equivalences is used.

Some firms did not report investment figures for one or more of the survey periods. Since the

total number of firms (and observations) in the sample is already at the lower limit, I tried

to keep as many observations in the sample as possible. Therefore I imputed investments for

firms with missing values by multiplying the total number of employees with industry and year

specific average investment intensities (investment per employee) obtained from the full survey

sample (full cross section) in each specific survey year. In order to construct a capital stock

from investment data, I use official producer price deflators for investment goods to deflate the

investments of firm i. Given the deflated investments for capital, I apply the perpetual inventory

6 For these calculations I used Table 81000-0103 and Table 81000-0101 from the German Statistical Office.7 If Zit and Yit are sales and value added of firm i in period t, and if Zj(i),t and Yj(i),t are sales and value addedaggregated over all firms of the same industry j(i) that firm i is operating in. Then the unknown value addedof firm i is approximated by Yit ' Zit · Yj(i),t/Zj(i),t.

9

method with constant, geometric depreciation to construct the capital stock. Accordingly, the

capital stock Kit of firm i in period t results from investment Ii,t−1 in the following way:

Kit = (1− δj(i))Ki,t−1 + Ii,t−1 (3)

with δj(i) denoting the industry-specific depreciation rates of capital stocks for firm i.8 Since no

information is available on the level of capital stocks, I construct initial capital stocks employing

the method proposed by Hall and Mairesse (1995). Under the assumption that investment

expenditures on capital goods have grown at a similar, constant average rate g in the past in

all firms, Equation (3) can be rewritten for period t = 1 (2000) by backward substitution in the

following way:9

K1 = I1 + (1− δ)I−1 + (1− δ)2I−2 + . . .

=∞∑s=0

I−s(1− δ)s

= I0

∞∑s=0

[1− δ1 + g

]s=

I1g + δ

(4)

In order to derive the initial capital stocks, assumptions about the pre-period growth rate g

of investments have to be made. Hereby, I assume an annual growth rate of approximately

5 percent (g = 0.05). Since there are time lags between the installation and the productive

contribution of capital goods, I employ the capital stock at the beginning of each period (or

at the end of the corresponding previous period) as a measure of capital input. In order to

8 I calculated the depreciation rates δj(i) by industries as the shares of capital consumption in net fixed assetsevaluated at replacement prices (time series 81000-0107 and 81000-0117 of the German Statistical Office). Theunweighted mean over all industries amounts to 4.8 percent with a maximum in NACE 72 (data processing)of 26.0 percent and a minimum in NACE 70 (real estate) with 2.3 percent.

9 In fact, the initial value of investment for firm i Ii,1 is replaced by the average of the observed values ofinvestment such that Ii,1 ' 1

T

∑Tt=1 Iit .

10

apply suited econometric techniques (system-GMM estimation), I consider only firms for which

consistent information is available for at least three consecutive periods.10

The resulting unbalanced sample consists of 700 firms with a total of 2371 observations.

The statistics of the total sample, the outsourcing firms and the non-outsourcing firms are

summarised in Table 1. The figures report the means and medians of the unweighted firms’

means over time. The majority of firms in the reference sample are small and medium–sized

firms with a median of 49 employees. The median value of the number of employees is higher

for firms involved in outsourcing (61 employees) than for non-outsourcing firms (38 employees).

The bottom rows of Table 1 report the means and medians of the firms’ (longitudinal) averages

of output and capital intensity (value added and capital per employee). At the median firm,

a workplace is equipped with capital worth e60,950. The median value added per employee is

e62,900. Again value added per employee and capital per employee is higher for the outsourcing

sub-sample compared with the non outsourcing sample. 11

The business process outsourcing variable , the main variable of interest in this analysis, is a

dummy variable indicating if firm i partly or completely outsources business processes to an

external service provider. The variable is time-invariant, taking the value one if the firm began

to outsource business processes in the year 2000 or before, with 2000 being the first year for

which data from the survey is available. The questionnaire of the ICT survey asked firms about

their outsourcing engagement in certain business activities, the starting year of this engagement

and the extend of their outsourcing (fully or partly). A firm belongs to the outsourcing group if

at least one of the following business processes is provided by an external service provider: (i) IT

services, (ii) Marketing, (iii) Procurement, (iv) Customer Services and (v) Sales and Distribution.

10 Since information about BPO is essential for the analysis conducted in this paper and BPO information wasonly collected in the 2007 survey, all firms included in the final sample were observed in the year 2007. Thepanel observations then reach back at least to the year 2002 and sometimes even further to the first wave ofthe ZEW ICT-survey in 2000.

11 The corresponding mean values are substantially higher than the median since some firms display very highvalues for both inputs and outputs per employee.

11

Taking a look at Table 2 reveals the decomposition of the sample by industry affiliation. One

can easily see that the greatest share of firms in the sample belongs to the metal and machine

construction industry. The smallest share of firms on the other hand belongs to the wholesale

trade industry.12 As the last two columns of Table 2 indicate, business process outsourcing is

extremely popular in the chemical industry. More than 61 percent of the firms belonging to

this industry are involved in BPO. The industry with the least involvement in business process

outsourcing is the electronic processing and telecommunication industry. Here, only 26 percent

of the firms outsource business processes to an external service provider. Overall, 43 percent of

all firms are outsourcing companies.

Table 3 shows descriptive statistics of labour productivity (value added per employee), separated

into firms with BPO and firms without BPO. The mean of labour productivity is substantially

higher for firms that are involved in outsourcing compared to those firms without any involve-

ment in BPO. Outsourcing firms reach a mean value for labour productivity of about e114,063

and non-outsourcing firms show a value of e74,208. According to a t-test the difference is

also highly significant at any conventional level. When calculating the logarithm of labour

productivity (the logarithm of productivity is later on used in the estimation procedures) and

comparing the means in the two sub-samples a t-test again shows that the mean of the logarithm

of labour productivity is significantly higher in the sample of firms with outsourcing activities.

5 Results

Tables 4 and 5 show the results for pooled OLS, random effects and system-GMM estimation.

The first two columns of Table 4 contain the results of the pooled OLS regression for a pure

productivity estimation with labour and capital inputs and productivity estimation. The

additional dummy variable indicates if firm i is involved in business process outsourcing. In

both estimations the labour and capital input coefficients are highly significant, reaching values

12 For a detailed description and composition of the sectors included in the survey, see Table 6. In the ZEWICT-survey banks and insurances are also surveyed, but since there is no output data available for those firms(only the balance sheet total and insurance premiums, respectively), these industries are excluded from theanalysis in this paper.

12

of about 0.96 for labour input and 0.08 (0.07) for capital input. The coefficient for BPO is also

positive and highly significant, reaching a value of 0.25. The pooled OLS estimates are possibly

biased because observations of the same firm in different years are considered as independent

and unobserved heterogeneity cannot be taken into account.

To account for this possible bias, I estimated a random effects model.13 The estimation results

are shown in the third column of Table 4. In contrast to the pooled OLS regression of column

two, the capital coefficient is significantly smaller and not significant at the usual significance

level any more. Additionally, the labour coefficient increased to 0.99. Moreover, the BPO

coefficient slightly increased to a value of 0.27 and is still highly significant.

The endogeneity problem of labour and capital is further addressed in the system-GMM

regressions. Here, the lagged endogenous variables are used as instruments. Labour and capital

are regarded as endogenous variables, the dummies for industry, time, and the location of

the firm (East or West Germany) are assumed to be exogenous. Besides, the BPO dummy

variable, which indicates if firm i has done business process outsourcing since at least the

year 2000 is assumed to be exogenous. System-GMM estimation results are presented in

5. I estimated a basic production function without the ‘BPO-input’ variable as a starting

point. The results for the labour and capital inputs are again highly significant, as well as the

coefficient for the dummy variable indicating if a firm is located in East or West Germany.

In contrast to the OLS and random effects regressions, the coefficient for labour is smaller

reaching a value of 0.81. However, the capital input coefficient increased to around 0.26, which

is considerably higher than 0.08 in the pooled OLS regression. Adding BPO as an input factor

to the production function estimation and focusing on all firms in the sample, I can find a

significant positive impact of business process outsourcing on productivity at the firm level. The

coefficient reaches a value of 0.21 and is highly significant. In this setup, the labour and capi-

tal input coefficients do not differ much compared to the basic setup of the productivity equation.

13 A fixed effect model would also be an appropriate alternative in this case. However, since the variable of maininterest is a time-invariant dummy, this variable would not be identified in a fixed effect model.

13

In a further step, I want to see if there are differences concerning the productivity effects of

business process outsourcing in the different sectors of the manufacturing or service industries.

The results of these two estimations are reported in columns three and four of 5. In both

settings the labour input coefficient is highly significant but the difference in absolute terms is

quite substantial. While the coefficient for labour is around 0.72 in the manufacturing industry,

it is almost 0.96 in the service sector. This result is not surprising, since manufacturing is

generally less labour intensive than the service sector. Consequently, the capital coefficient is

higher for manufacturing than for services, albeit in neither estimation the coefficient achieves

significance. The most interesting finding is that in both estimations the impact of business

process outsourcing is positive and highly significant (at the five percent level in manufacturing

and at the one percent level in services). In absolute terms, the impact seems to be more

pronounced in the service sector (0.33 compared to 0.27).

In all system-GMM estimations, there is significant first order correlation (of the first differenced

residuals) and no second order correlation at the usual levels. This result indicates the validity of

the applied instruments. This is further undermined by the Hansen J -Test of overidentification.

In all settings the null hypothesis could not be rejected.

6 Concluding Remarks

The aim of this paper is to analyse the impact of business process outsourcing on firm-level

productivity. An augmented production function approach is used which takes account of

firms’ BPO activities. For the empirical analysis system-GMM estimation is applied for a

panel data set of German firms. The results show that business process outsourcing has a

considerably positive and significant effect on firm-level productivity. This result holds for firms

of the manufacturing as well as for firms from the services sector. The coefficient indicating if

a firm is involved in BPO is again highly positive and significant in separate estimations for

manufacturing firms and service firms.

14

The outsourcing of business processes to external providers seems to be a good choice for firms.

The allows managers to focus on the core business of the firm. Moreover, the qualified and

experienced work of the service provider and the cost savings finally result in an improved

business performance. This research, however, is still at the beginning. Further research efforts

are necessary and should focus for example on a decomposition of labour input by qualification

and age. Moreover, interactions between BPO and specific input factors could be taken into

account.

15

References

Abraham, K. and Taylor, S. (1996). Firms’ Use of Outside Contractors: Theory and Evidence,Journal of Labor Economics 14(3): 394–424.

Abramovsky, L. and Griffith, R. (2006). Outsourcing and Offshoring of Business Services: HowImportant is ICT?, Journal of the European Economic Association 4(2–3): 594–601.

Amiti, M. and Wei, S. (2005). Fear of Service Outsourcing: Is it Justified?, Economic Policy20(42): 307–347.

Antonietti, R. and Cainelli, G. (2007). Spatial Agglomeration, Technology and Outsourcingof Knowledge Intensive Business Services: Empirical Insights from Italy, Discussion Paper79.2007, Fondazione Eni Enrico Mattei, Milan.

Antràs, P., Garicano, L. and Rossi-Hansberg, E. (2006). Offshoring in a Knowledge Economy,Quarterly Journal of Economics 121(1): 31–77.

Arellano, M. and Bond, S. (1991). Some Tests of Specification for Panel Data: Monte Carlo Evi-dence and an Application to Employment Equations, Review of Economic Studies 58(2): 277–297.

Arellano, M. and Bover, O. (1995). Another Look at the Instrumental Variable Estimation ofError-Components Models, Journal of Econometrics 68(1): 29–51.

Barthêlemy, J. and Geyer, D. (2001). IT Outsourcing: Evidence from France and Germany,European Management Journal 19(2): 195–202.

Barthêlemy, J. and Geyer, D. (2004). The Determinants of Total IT Outsourcing: An Empir-ical Investigation of French and German Firms, Journal of Computer Information Systems44(3): 91–97.

Barthélemy, J. and Geyer, D. (2005). An Empirical Investigation of IT Outsourcing versusQuasi-Outsourcing in France and Germany, Information & Management 42(4): 533–542.

Bertschek, I. and Müller, M. (2006). Productivity Effects of IT-Outsourcing: SemiparametricEvidence for German Companies, in S. Sperrlich, W. Härdle and G. Adinli (eds), The Art ofSemiparametrics, Physica-Verlag, Heidelberg, pp. 130–154.

Blundell, R. and Bond, S. (1998). Initial Conditions and Moment Restrictions in Dynamic PanelData Models, Journal of Econometrics 87(1): 115–143.

Dibbern, J., Goles, T., Hirschheim, R. and Jayatilaka, B. (2004). Information Systems Out-sourcing: A Survey and Analysis of the Literature, DATA BASE for Advances in InformationSystems 35(4): 6–102.

Fixler, D. and Siegel, D. (1999). Outsourcing and Productivity Growth in Services, StructuralChange and Economic Dynamics 10: 177–194.

Girma, S. and Görg, H. (2004). Outsourcing, Foreign Ownership, and Productivity: Evidencefrom UK Establishment-level Data, Review of International Economics 12(5): 817–832.

Görg, H. and Hanley, A. (2005). International Outsourcing and Productivity: Evidence from theIrish Electronics Industry, North American Journal of Economics and Finance 16(2): 255–269.

Görg, H., Hanley, A. and Strobl, E. (2008). Productivity Effects of International Outsourcing:Evidence from Plant-Level Data, Canadian Journal of Economics 41(2): 670–688.

16

Grossman, G. and Helpman, E. (2003). Outsourcing versus FDI in Industry Equilibrium, Journalof the European Economic Association 1(2-3): 317–327.

Grossman, G. and Helpman, E. (2005). Outsourcing in a Global Economy, Review of EconomicStudies 72(250): 135–159.

Hall, B. and Mairesse, J. (1995). Exploring the Relationship Between R&D and Productivity inFrench Manufacturing Firms, Journal of Econometrics 65(1): 263–293.

Hempell, T. (2006). Computers and Productivity - How Firms Make a General Purpose Tech-nology Work, Vol. 33, ZEW Economic Studies edn, Physica-Verlag, Heidelberg.

Heshmati, A. (2003). Productivity Growth, Efficiency and Outsourcing in Manufacturing andService Industries, Journal of Economic Surveys 17(1): 79–112.

Hijzen, A. (2007). International Outsourcing, Technological Change, and Wage Inequality, Re-view of International Economics 15(1): 188–205.

Jiang, B. and Qureshi, A. (2006). Research on Outsourcing Results: Current Literature andFuture Opportunities, Management Decision 44(1): 44–55.

Loh, L. and Venkatraman, N. (1992). Determinants of Information Technology Outsourcing: ACross-Sectional Analysis, Journal of Management Information Systems 9(1): 7–24.

Maliranta, M., Rouvinen, P. and Airaksinen, A. (2008). IT Outsourcing in Finnish Business,Discussion Paper 1140, Research Institute of the Finnish Economy.

Merino, F. and Rodríguez Rodríguez, D. (2007). Business Services Outsourcing by ManufacturingFirms, Industrial and Corporate Change 16(6): 1147–1173.

Mukherjee, S. (2007). Business Process Outsourcing and India, Working paper, Baroda, Gujarat,India.

Ohnemus, J. (2007). Does IT Outsourcing Increase Firm Success? An Empirical Assessmentusing Firm-Level Data, Discussion Paper 07-087, ZEW.

Olsen, K. (2006). Productivity Impacts of Offshoring and Outsourcing: A Review, OECD Sci-ence, Technology and Industry Working Paper 2006/1, OECD, Paris.

Roodman, D. (2006). How to Do xtabond2: An Introduction to Difference and System GMM inStata, Working Paper 103, Center for Global Development.

Sako, M. (2006). Outsourcing and Offshoring Implications for Productivity of Business Services,Oxford Review of Economic Policy 22(4): 499–512.

Sharma, C. (2004). BPO Business: It’s A Win-Win Situation, Financial World 3(276).

Siegel, D. and Griliches, Z. (1992). Purchased Services, Outsourcing, Computers, and Produc-tivity in Manufacturing, in Z. Griliches (ed.), Output Measurement in the Service Sectors,University of Chicago Press, Chicago, IL, pp. 429–460.

Windmeijer, F. (2005). A Finite Sample Correction for the Variance of Linear Efficient Two-stepGMM Estimators, Journal of Econometrics 126: 25–51.

ZEW (2007). Internet Economy Continuously on the Rise, ICT-Report, ZEW, Mannheim.

17

Appendix

A – Data

Output (Y): Firms’ value added is used as output measure. Because of data restrictions

only data of sales is available. I constructed firm-specific and deflated value added figures by

multiplying sales with the shares of real value added in nominal gross output terms at the

NACE two–digit industry level.

Employment (L): For labour input, the number of employees in full-time equivalences is used.

Besides the total number of employees, the firms were asked about their share of part-time

employees. I accounted for that share, assuming that part-time employees work on average half

the time of full-time employees.

Capital Stock (K): The capital stock variable is constructed out of deflated investment

information available for each firm and year. I apply the perpetual inventory method to achieve

capital stock data for each firm (and year).

Business Process Outsourcing (BPO): The business process outsourcing variable is a

dummy which takes on the value one if firm i started to partly or completely outsource (al least)

one of the following business processes before or in the year 2000: (i) IT services, (ii) Marketing,

(iii) Procurement, (iv) Customer Services and (v) Sales and Distribution.

East or West Germany: Dummy variable indicating if the location of firm i is in East or

West Germany.

18

B – Figures and Tables

Figure 1: Share of intermediate inputs from “other business activities” sector in the total production value inGermany (1995-2005)

5,455,37

5,08

5,71

5,975,99

6,196,33 6,31

6,12

6,38

4

5

6

7

8

1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005

perc

enta

ge s

hare

share of intermediate input plus imports from "other business activities" sector at total production value

share of intermediate inputs from "other business activities" sector at total production value

Source: Based on input-output tables provided by the Germany Statistical Office and authors’ calculations.

Figure 2: Share of intermediate inputs from “corporate service sector”∗ in the total production value inGermany (1995-2005)

7,30

7,03

7,277,37

7,26

6,97

6,85

6,54

5,79

6,066,07

5

6

7

8

1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005

perc

enta

ge s

hare

share of intermediate input plus imports from the corporate service sector* at total production value

share of intermediate inputs from the corporate service sector* at total production value

Note: ∗The “corporate service sector” comprises the sectors “computer and related activities” (NACE 72), “research anddevelopment” (NACE 73) and “other business activities” (NACE 74).Source: Based on input-output tables provided by the Germany Statistical Office and authors’ calculations.

19

Figure 3: Share of intermediate service inputs at total value added in Germany (1978-2005)

0

5

10

15

20

25

30

35

40

1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004

perc

enta

ge s

hare

Service Industry Manufacturing

Source: EU-Klems 2008 and authors’ calculations.

Figure 4: Share of firms outsourcing business processes in Germany 2007

54

36

23

8

4

4

4

59

44

28

12

7

4

4

0 10 20 30 40 50 60 70

Overall

Accounting

Human Resources

Marketing

Procurement

Customer Service

Sales and Distribution

percentage share

Manufacturing

Service Industries

Source: ZEW ICT-survey, first quarter 2007.

20

Table 1: Descriptive Statistics

Industry All Outsourcing Non-OutsourcingFirms Firms Firms

Mean Median Mean Median Mean Median

sales 54071.27 7008.61 82827.79 10866.67 32807.15 4833.33value added 23379.33 3270.91 34541.41 4932.43 15125.47 2496.82employees 223.72 48.75 305.41 60.83 163.31 38.33capital 153961.96 2933.57 265760.78 4334.97 71291.86 2003.06

log value added 8.08 8.03 8.53 8.44 7.75 7.65log employees 3.92 3.83 4.18 4.08 3.73 3.53log capital 8.13 7.97 8.55 8.36 7.82 7.60

value added p. employee 91.15 62.90 114.06 67.20 74.21 60.89capital p. employee 2261.49 60.95 3215.86 68.89 1555.77 57.61

Number of firms 701 298 403

Note: The figures are calculated at the means and medians of the unweighted firms’ means over time. Monetary valuesare in 1,000 Euros in prices of 2000. Source: ZEW ICT-survey.

Table 2: Share of Observations by Industry

Industry All thereof ...Firms Outsourcing Non-Outsourcing

consumer goods 7.9 45.5 54.6chemical industry 6.3 61.4 38.6other raw materials 7.6 39.6 60.4metal and machine construction 13.4 41.5 58.5electrical engineering 8.7 32.8 67.2precision instruments 9.3 33.9 66.2automobile 5.3 56.8 43.2wholesale trade 4.0 57.1 42.9retail trade 7.4 55.8 44.2transport and postal serv. 7.0 53.1 46.9electronic process. and telecom. 10.6 25.7 74.3technical services 7.4 34.6 65.4other business-related serv. 5.1 41.7 58.3

Total 100.0 42.6 57.4

Source: ZEW ICT-survey.

21

Table 3: Comparison of Mean Labour Productivity

QuantileN 10% 50% 90% Mean Std.Dev.

Outsourcing FirmsLabour Productivity 298 40 67 227 114.063 130.931Non-Outsourcing FirmsLabour Productivity 403 27 61 146 74.208 54.417

t-test on the equality of the means of Labour ProductivityH0: mean(w/ BPO) - mean(w/o BPO) = diff = 0 → t = 5.5028H1: diff 6= 0 → [ p > |t | ] = 0.0000

t-test on the equality of the means of log Labour ProductivityH0: mean(w/ BPO) - mean(w/o BPO) = diff = 0 → t = 6.5412H1: diff 6= 0 → [ p > |t | ] = 0.0000

Note: Labour productivity is value added per employee in 1,000 Euro in prices of 2000. It is calculated at the means andmedians of the unweighted firms’ means over time. Source: ZEW ICT-survey.

Table 4: Estimation Results – OLS and Random Effects

(1) (2) (3)

OLS OLS Random Effects

Labour 0.9641*** 0.9577*** 0.9946***(0.0233) (0.0224) (0.0642)

Capital 0.0778*** 0.0723*** 0.0471(0.0180) (0.0170) (0.0539)

BPO –– 0.2571*** 0.2690***(0.0466) (0.0491)

East Germany -0.2346*** -0.2131*** -0.1928***(0.0624) (0.0612) (0.0597)

Constant 3.4538*** 3.4404*** 4.6081***(0.2179) (0.2300) (0.3410)

R2 0.8418 0.8466 0.8469Number of observations 2371 2371 1671Number of firms 700

Note: *,** and *** indicate significance at the 10%, 5% and 1% level respectively. Robust standard errors are reported.Source: ZEW ICT-survey and own calculations.

22

Table 5: Estimation Results – SYS-GMM

(1) (2) (3) (4)

Basic All Manufact. Services

Labour 0.8060*** 0.8423*** 0.7238*** 0.9580***(0.1543) (0.1426) (0.1932) (0.1462)

Capital 0.2569** 0.2240* 0.1010 0.0599(0.1254) (0.1340) (0.2062) (0.0841)

BPO –– 0.2053** 0.2721** 0.3281***(0.0858) (0.1075) (0.1039)

East Germany -0.1786** -0.1634** -0.2506** -0.1808(0.0880) (0.0810) (0.0982) (0.1117)

Constant 2.9602*** 3.0126*** 4.4526*** 3.9048***(1.0392) (1.0277) (1.4946) (0.7575)

AR(1) (p-values) 0.0000 0.0000 0.0000 0.0010AR(2) (p-values) 0.5680 0.5401 0.2748 0.5633Hansen J -Test (p-values) 0.5600 0.5647 0.3432 0.6139R2 0.8229 0.8334 0.8663 0.7995Number of instruments 63 64 40 36

Number of observations 2371 2371 1383 988Number of firms 700 700 409 291

Note: *,** and *** indicate significance at the 10%, 5% and 1% level respectively. Robust standard errors are reported.Source: ZEW ICT-survey and own calculations.

23

Table 6: Industry Classification

Industry Explanation NACE

consumer goodsmanufacture of food products, beverages and tobacco 15-16manufacture of textiles and textile products 17-18manufacturing of leather and leather products 19manufacture of wood and wood products 20manufacturing of pulp, paper and paper products; publishing and printing 21-22manufacturing n.e.c. 36-37

chemical industrymanufacture of coke, refined petroleum products and nuclear fuel 23manufacture of chemicals, chemical products and man-made fibres 24

other raw materialsmanufacture of rubber and plastic products 25manufacture of non-metallic mineral products 26manufacture of basic metal 27

metal and machine constructionmanufacture of fabricated metal products (except machinery and equipment) 28manufacture of machinery and equipment n.e.c. 29

electrical engineeringmanufacture of office machinery and computers 30manufacture of electrical machinery and apparatus n.e.c. 31manufacture of radio, television and communication equipment and apparatus 32

precision instrumentsmanufacture of medical, precision and optical instruments, watches and clocks 33

automobilemanufacturing of transport equipment 34-35

wholesale tradewholesale trade and commission trade (except of motor vehicles and motorcycles) 51

retail tradesale, maintenance and repair of motor vehicles and motorcycles; retail sale of automotive fuel 50retail trade (except of motor vehicles and motorcycles), repair of personal and household goods 52

transportation and postal servicesland transport, transport via pipeline 60water transport 61air transport 62supporting and auxiliary transport activities; activities of travel agencies 63post and courier activities 64.1

electronic processing and telecommunicationcomputer and related activities 72telecommunications 64.2

technical servicesresearch and development 73architectural and engineering activities and related technical consultancy 74.2technical testing and analysis 74.3

other business-related servicesreal estate activities 70renting of machinery without operator and of personal and household goods 71legal, accounting, book keeping and auditing activities; tax consultancy; market research andpublic opinion pools; business and management consultancy; holdings

74.1

advertising 74.4labour recruitment and provision of personnel 74.5investigation and security services 74.6industrial cleaning 74.7miscellaneous business activities n.e.c. 74.8sewage and refuse disposal, sanitation and similar activities 90

24